Vehicle Detection and Tracking Project¶
Read training and testing data¶
In [1]:
import numpy as np
import cv2
import glob
import matplotlib.pyplot as plt
from sklearn.utils import shuffle
%matplotlib inline
vehicle_images_path = shuffle(glob.glob('./vehicles/*/*.png'))
non_vehicle_images_path = shuffle(glob.glob('./non-vehicles/*/*.png'))
sample_vehicle_images = []
sample_non_vehicle_images = []
fig1 = plt.figure()
for i in range(5):
sample_vehicle_images.append(cv2.imread(vehicle_images_path[i]))
fig1.add_subplot(1, 5, i+1)
if i == 0:
plt.title('Car image samples:', fontsize=28)
plt.imshow(sample_vehicle_images[i])
fig1.set_size_inches(np.array(fig1.get_size_inches()) * 5)
plt.show()
fig2 = plt.figure()
for i in range(5):
sample_non_vehicle_images.append(cv2.imread(non_vehicle_images_path[i]))
fig2.add_subplot(1, 5, i+1)
if i == 0:
plt.title('Non-Car image samples:', fontsize=28)
plt.imshow(sample_non_vehicle_images[i])
fig2.set_size_inches(np.array(fig2.get_size_inches()) * 5)
plt.show()
Implement HOG feature extraction¶
In [2]:
from skimage.feature import hog
def get_hog_features(img, orient, pix_per_cell, cell_per_block,
vis=False, feature_vec=True):
# Call with two outputs if vis==True
if vis == True:
features, hog_image = hog(img, orientations=orient,
pixels_per_cell=(pix_per_cell, pix_per_cell),
cells_per_block=(cell_per_block, cell_per_block),
block_norm= 'L2-Hys',
transform_sqrt=False,
visualise=vis, feature_vector=feature_vec)
return features, hog_image
# Otherwise call with one output
else:
features = hog(img, orientations=orient,
pixels_per_cell=(pix_per_cell, pix_per_cell),
cells_per_block=(cell_per_block, cell_per_block),
block_norm= 'L2-Hys',
transform_sqrt=False,
visualise=vis, feature_vector=feature_vec)
return features
image = sample_vehicle_images[0]
gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
features, hog_image = get_hog_features(
gray, orient= 9, pix_per_cell= 8, cell_per_block= 2, vis=True, feature_vec=False)
# Plot the car examples
car_hog_fig = plt.figure()
plt.subplot(121)
plt.imshow(image, cmap='gray')
plt.title('Example Car Image')
plt.subplot(122)
plt.imshow(hog_image, cmap='gray')
plt.title('Car HOG Visualization')
image = sample_non_vehicle_images[0]
gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
features, hog_image = get_hog_features(
gray, orient= 9, pix_per_cell= 8, cell_per_block= 2, vis=True, feature_vec=False)
# Plot the non-car examples
non_car_hog_fig = plt.figure()
plt.subplot(121)
plt.imshow(image, cmap='gray')
plt.title('Example Non-car Image')
plt.subplot(122)
plt.imshow(hog_image, cmap='gray')
plt.title('Non-car HOG Visualization')
Out[2]:
Implement color feature extraction¶
In [3]:
def bin_spatial(img, size=(32, 32)):
color1 = cv2.resize(img[:,:,0], size).ravel()
color2 = cv2.resize(img[:,:,1], size).ravel()
color3 = cv2.resize(img[:,:,2], size).ravel()
return np.hstack((color1, color2, color3))
def color_hist(img, nbins=32): #bins_range=(0, 256)
# Compute the histogram of the color channels separately
channel1_hist = np.histogram(img[:,:,0], bins=nbins)
channel2_hist = np.histogram(img[:,:,1], bins=nbins)
channel3_hist = np.histogram(img[:,:,2], bins=nbins)
# Concatenate the histograms into a single feature vector
hist_features = np.concatenate((channel1_hist[0], channel2_hist[0], channel3_hist[0]))
# Return the individual histograms, bin_centers and feature vector
return hist_features
image = sample_vehicle_images[0]
feature_vec = bin_spatial(image)
color_hist_feature = color_hist(image)
# Plot the car examples
car_color_fig = plt.figure()
car_color_fig.set_size_inches(np.array(car_hog_fig.get_size_inches()) * 3)
plt.subplot(131)
plt.imshow(image, cmap='gray')
plt.title('Example Car Image', fontsize=28)
plt.subplot(132)
plt.plot(feature_vec)
plt.title('Example Car Spatially Binned Features', fontsize=28)
plt.subplot(133)
plt.plot(color_hist_feature)
plt.title('Example Color Histogram Features', fontsize=28)
image = sample_non_vehicle_images[0]
feature_vec = bin_spatial(image)
color_hist_feature = color_hist(image)
# Plot the non-car examples
non_car_color_fig = plt.figure()
non_car_color_fig.set_size_inches(np.array(car_hog_fig.get_size_inches()) * 3)
plt.subplot(131)
plt.imshow(image, cmap='gray')
plt.title('Example Non-car Image', fontsize=28)
plt.subplot(132)
plt.plot(feature_vec)
plt.title('Example Non-car Spatially Binned Features', fontsize=28)
plt.subplot(133)
plt.plot(color_hist_feature)
plt.title('Example Non-car color Histogram Features', fontsize=28)
Out[3]:
Concatnate binned color feature and histograms of color and HOG¶
In [4]:
def concatnate_features(hog_features, image, size=(32, 32), nbins=32):
spatial_features = bin_spatial(image, size=size)
hist_features = color_hist(image, nbins=nbins)
features = np.concatenate([spatial_features, hist_features, hog_features])
return features
image = sample_vehicle_images[0]
gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
hog_features = get_hog_features(
gray, orient= 9, pix_per_cell= 8, cell_per_block= 2, vis=False, feature_vec=True)
concatnated_features = concatnate_features(hog_features, image)
# Plot the car examples
car_color_fig = plt.figure()
car_color_fig.set_size_inches(np.array(car_hog_fig.get_size_inches()) * 2)
plt.subplot(121)
plt.imshow(image, cmap='gray')
plt.title('Example Car Image', fontsize=28)
plt.subplot(122)
plt.plot(concatnated_features)
plt.title('Example Car Concatnated Features', fontsize=28)
image = sample_non_vehicle_images[0]
gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
hog_features = get_hog_features(
gray, orient= 9, pix_per_cell= 8, cell_per_block= 2, vis=False, feature_vec=False).ravel()
concatnated_features = concatnate_features(hog_features, image)
# Plot the non-car examples
non_car_color_fig = plt.figure()
non_car_color_fig.set_size_inches(np.array(car_hog_fig.get_size_inches()) * 2)
plt.subplot(121)
plt.imshow(image, cmap='gray')
plt.title('Example Non-Car Image', fontsize=28)
plt.subplot(122)
plt.plot(concatnated_features)
plt.title('Example Non-Car Concatnated Features', fontsize=28)
Out[4]:
Normalize feature and randomize data set¶
In [5]:
vehicle_images = []
non_vehicle_images = []
for path in vehicle_images_path:
vehicle_images.append(cv2.imread(path))
for path in non_vehicle_images_path:
non_vehicle_images.append(cv2.imread(path))
In [6]:
def extract_features(imgs, color_space='RGB', spatial_size=(32, 32),
hist_bins=32, orient=9,
pix_per_cell=8, cell_per_block=2, hog_channel=0,
spatial_feat=True, hist_feat=True, hog_feat=True):
# Create a list to append feature vectors to
features = []
# Iterate through the list of images
for image in imgs:
file_features = []
feature_image = image
if color_space != 'RGB':
if color_space == 'HSV':
feature_image = cv2.cvtColor(image, cv2.COLOR_RGB2HSV)
elif color_space == 'LUV':
feature_image = cv2.cvtColor(image, cv2.COLOR_RGB2LUV)
elif color_space == 'HLS':
feature_image = cv2.cvtColor(image, cv2.COLOR_RGB2HLS)
elif color_space == 'YUV':
feature_image = cv2.cvtColor(image, cv2.COLOR_RGB2YUV)
elif color_space == 'YCrCb':
feature_image = cv2.cvtColor(image, cv2.COLOR_RGB2YCrCb)
else: feature_image = np.copy(image)
if spatial_feat == True:
spatial_features = bin_spatial(feature_image, size=spatial_size)
file_features.append(spatial_features)
if hist_feat == True:
# Apply color_hist()
hist_features = color_hist(feature_image, nbins=hist_bins)
file_features.append(hist_features)
if hog_feat == True:
# Call get_hog_features() with vis=False, feature_vec=True
if hog_channel == 'ALL':
hog_features = []
for channel in range(feature_image.shape[2]):
hog_features.append(get_hog_features(feature_image[:,:,channel],
orient, pix_per_cell, cell_per_block,
vis=False, feature_vec=True))
hog_features = np.ravel(hog_features)
else:
hog_features = get_hog_features(feature_image[:,:,hog_channel], orient,
pix_per_cell, cell_per_block, vis=False, feature_vec=True)
# Append the new feature vector to the features list
file_features.append(hog_features)
features.append(np.concatenate(file_features))
# Return list of feature vectors
return features
color_space = 'HSV' # Can be RGB, HSV, LUV, HLS, YUV, YCrCb
orient = 9 # HOG orientations
pix_per_cell = 8 # HOG pixels per cell
cell_per_block = 4 # HOG cells per block
hog_channel = 1 # Can be 0, 1, 2, or "ALL"
spatial_size = (32, 32) # Spatial binning dimensions
hist_bins = 32 # Number of histogram bins
spatial_feat = True # Spatial features on or off
hist_feat = True # Histogram features on or off
hog_feat = True # HOG features on or off
car_features = extract_features(vehicle_images, color_space=color_space,
spatial_size=spatial_size, hist_bins=hist_bins,
orient=orient, pix_per_cell=pix_per_cell,
cell_per_block=cell_per_block,
hog_channel=hog_channel, spatial_feat=spatial_feat,
hist_feat=hist_feat, hog_feat=hog_feat)
notcar_features = extract_features(non_vehicle_images, color_space=color_space,
spatial_size=spatial_size, hist_bins=hist_bins,
orient=orient, pix_per_cell=pix_per_cell,
cell_per_block=cell_per_block,
hog_channel=hog_channel, spatial_feat=spatial_feat,
hist_feat=hist_feat, hog_feat=hog_feat)
# Create an array stack of feature vectors
X = np.vstack((car_features, notcar_features)).astype(np.float64)
# Define the labels vector
y = np.hstack((np.ones(len(car_features)), np.zeros(len(notcar_features))))
from sklearn.cross_validation import train_test_split
from sklearn.preprocessing import StandardScaler
# Split up data into randomized training and test sets
rand_state = np.random.randint(0, 100)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=rand_state)
# Fit a per-column scaler
X_scaler = StandardScaler().fit(X_train)
# Apply the scaler to X
X_train = X_scaler.transform(X_train)
X_test = X_scaler.transform(X_test)
In [7]:
image = sample_vehicle_images[0]
feature_image = cv2.cvtColor(image, cv2.COLOR_RGB2HSV)
hog_features = get_hog_features(
feature_image[:,:,1], orient, pix_per_cell, cell_per_block, vis=False, feature_vec=True)
concatnated_features = concatnate_features(hog_features, image)
normalized_concatnated_features = X_scaler.transform(np.array([concatnated_features]))
# Plot the car examples
car_color_fig = plt.figure()
car_color_fig.set_size_inches(np.array(car_hog_fig.get_size_inches()) * 2)
plt.subplot(121)
plt.imshow(image, cmap='gray')
plt.title('Example Car Image', fontsize=28)
plt.subplot(122)
plt.plot(normalized_concatnated_features[0])
plt.title('Example Car Normalized Concatnated Features', fontsize=28)
image = sample_non_vehicle_images[0]
gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
hog_features = get_hog_features(
feature_image[:,:,hog_channel], orient, pix_per_cell, cell_per_block, vis=False, feature_vec=True)
concatnated_features = concatnate_features(hog_features, image)
normalized_concatnated_features = X_scaler.transform(np.array([concatnated_features]))
# Plot the non-car examples
non_car_color_fig = plt.figure()
non_car_color_fig.set_size_inches(np.array(car_hog_fig.get_size_inches()) * 2)
plt.subplot(121)
plt.imshow(image, cmap='gray')
plt.title('Example Non-Car Image', fontsize=28)
plt.subplot(122)
plt.plot(normalized_concatnated_features[0])
plt.title('Example Non-Car Normalized Concatnated Features', fontsize=28)
Out[7]:
Train linear SVM classifier¶
In [8]:
import time
from sklearn.svm import LinearSVC
# Use a linear SVC
svc = LinearSVC()
# Check the training time for the SVC
t=time.time()
svc.fit(X_train, y_train)
t2 = time.time()
print(round(t2-t, 2), 'Seconds to train SVC...')
# Check the score of the SVC
print('Test Accuracy of SVC = ', round(svc.score(X_test, y_test), 4))
Implement sliding window and search vechicles¶
In [9]:
def find_cars(img, ystart, ystop, scale, svc, X_scaler, orient, pix_per_cell, cell_per_block, spatial_size, hist_bins):
# Here we slightly from the code in the class, we cut the x from 450, since we realize the left side is uselss.
# Besides, the relative speed from the left side is much faster than right, it takes more false positive cases.
x_start = 450
bboxes = []
draw_img = np.copy(img)
img_tosearch = img[ystart:ystop,x_start:,:]
ctrans_tosearch = cv2.cvtColor(img_tosearch, cv2.COLOR_RGB2HSV)
if scale != 1:
imshape = ctrans_tosearch.shape
ctrans_tosearch = cv2.resize(ctrans_tosearch, (np.int(imshape[1]/scale), np.int(imshape[0]/scale)))
# Define blocks and steps as above
nxblocks = (ctrans_tosearch.shape[1] // pix_per_cell) - cell_per_block + 1
nyblocks = (ctrans_tosearch.shape[0] // pix_per_cell) - cell_per_block + 1
nfeat_per_block = orient*cell_per_block**2
# 64 was the orginal sampling rate, with 8 cells and 8 pix per cell
window = 64
nblocks_per_window = (window // pix_per_cell) - cell_per_block + 1
cells_per_step = 2 # Instead of overlap, define how many cells to step
nxsteps = (nxblocks - nblocks_per_window) // cells_per_step + 1
nysteps = (nyblocks - nblocks_per_window) // cells_per_step + 1
# Compute individual channel HOG features for the entire image
hog = get_hog_features(
ctrans_tosearch[:,:,hog_channel], orient, pix_per_cell, cell_per_block, vis=False, feature_vec=False)
for xb in range(nxsteps):
for yb in range(nysteps):
ypos = yb*cells_per_step
xpos = xb*cells_per_step
# Extract HOG for this patch
hog_feat = hog[ypos:ypos+nblocks_per_window, xpos:xpos+nblocks_per_window].ravel()
hog_features = np.array(hog_feat)
xleft = xpos*pix_per_cell
ytop = ypos*pix_per_cell
# Extract the image patch
subimg = cv2.resize(ctrans_tosearch[ytop:ytop+window, xleft:xleft+window], (64,64))
# Get color features
spatial_features = bin_spatial(subimg, size=spatial_size)
hist_features = color_hist(subimg, nbins=hist_bins)
# Scale features and make a prediction
test_features = X_scaler.transform(np.hstack((spatial_features, hist_features, hog_features)).reshape(1, -1))
#test_features = X_scaler.transform(np.hstack((shape_feat, hist_feat)).reshape(1, -1))
test_prediction = svc.predict(test_features)
if test_prediction == 1:
xbox_left = np.int(xleft*scale)
ytop_draw = np.int(ytop*scale)
win_draw = np.int(window*scale)
bboxes.append(((xbox_left+x_start, ytop_draw+ystart),(xbox_left+x_start+win_draw,ytop_draw+win_draw+ystart)))
cv2.rectangle(draw_img,(xbox_left, ytop_draw+ystart),(xbox_left+win_draw,ytop_draw+win_draw+ystart),(255,0,0),6)
return draw_img, bboxes
# We have an additional parameter nrounds, here, means how many sub space we will devide the search area.
def find_car_with_multiple_scales(img, ystart, ystop, scale, svc, X_scaler, orient, pix_per_cell, cell_per_block, spatial_size, hist_bins, nrounds=3):
draw_img = np.copy(img)
# We multiplify 3 here, makes each search area 50% overlap.
interval_per_round = int((ystop-ystart)/nrounds*1.5)
current_start = ystart
bboxes = []
for n in range(nrounds):
scale *= 1.5
_, intermediate_bboxes = find_cars(img, current_start, min(ystop, current_start+interval_per_round), scale, svc, X_scaler, orient, pix_per_cell, cell_per_block, spatial_size, hist_bins)
bboxes += [intermediate_bboxes]
current_start += int((ystop-ystart)/nrounds*(0.5 + n))
for bboxes_in_cetrain_size in bboxes:
for bbox in bboxes_in_cetrain_size:
cv2.rectangle(draw_img,bbox[0],bbox[1],(255,0,0),6)
return draw_img, bboxes
#img = cv2.imread('./test_images/test1.jpg')
img = cv2.imread('./video_frame_14.jpg')
ystart = 350
ystop = 650
scale = 1.2
nrounds=2
out_img, _ = find_car_with_multiple_scales(img, ystart, ystop, scale, svc, X_scaler, orient, pix_per_cell, cell_per_block, spatial_size, hist_bins, nrounds)
plt.imshow(cv2.cvtColor(out_img, cv2.COLOR_BGR2RGB))
Out[9]:
Create heatmap and estimate bounding boxes¶
In [10]:
from scipy.ndimage.measurements import label
def add_heat(heatmap, bbox_list, weight=1):
# Iterate through list of bboxes
for box in bbox_list:
# Add += 1 for all pixels inside each bbox
# Assuming each "box" takes the form ((x1, y1), (x2, y2))
heatmap[box[0][1]:box[1][1], box[0][0]:box[1][0]] += weight
# Return updated heatmap
return heatmap# Iterate through list of bboxes
def apply_threshold(heatmap, threshold):
# Zero out pixels below the threshold
heatmap[heatmap <= threshold] = 0
heatmap[heatmap > threshold] = 1
# Return thresholded map
return heatmap
def draw_labeled_bboxes(img, labels):
# Iterate through all detected cars
bboxes = []
for car_number in range(1, labels[1]+1):
# Find pixels with each car_number label value
nonzero = (labels[0] == car_number).nonzero()
# Identify x and y values of those pixels
nonzeroy = np.array(nonzero[0])
nonzerox = np.array(nonzero[1])
# Define a bounding box based on min/max x and y
bbox = ((np.min(nonzerox), np.min(nonzeroy)), (np.max(nonzerox), np.max(nonzeroy)))
if min(bbox[1][0] - bbox[0][0],bbox[1][1] - bbox[0][1]) > 60:
bboxes.append(bbox)
# Draw the box on the image
cv2.rectangle(img, bbox[0], bbox[1], (0,0,255), 6)
# Return the image
return img, bboxes
class CarsDetector():
def __init__(self):
self.memory_size = 15
self.recent_bboxes = []
self.heatmap = None
self.processed_image = None
def detect_car(self, img):
current_final_bboxes = []
self.processed_image = np.copy(img)
_, bboxes = find_car_with_multiple_scales(img, ystart, ystop, scale, svc, X_scaler, orient, pix_per_cell, cell_per_block, spatial_size, hist_bins, nrounds)
self.recent_bboxes.append(bboxes)
if len(self.recent_bboxes) > self.memory_size:
self.recent_bboxes = self.recent_bboxes[1:]
for i in range(nrounds):
heat = np.zeros_like(img[:,:,0]).astype(np.float)
# We only count heat map for same scale of bboxes.
weight = 1
for bboxes_in_one_frame in self.recent_bboxes:
heat = add_heat(heat*0.6,bboxes_in_one_frame[i], weight)
# Make sure first several frames will not have too big treshold
if len(self.recent_bboxes) > 5:
heat_current = apply_threshold(heat, 3)
else:
heat_current = apply_threshold(heat, 2)
heatmap_current = np.clip(heat_current, 0, 255)
labels = label(heatmap_current)
_, temp_bboxes = draw_labeled_bboxes(self.processed_image, labels)
#heat_for_future_use = apply_threshold(heat, 1.8)
#heatmap_for_future_use = np.clip(heat_for_future_use, 0, 255)
#labels = label(heatmap_for_future_use)
#_, temp_bboxes = draw_labeled_bboxes(self.processed_image, labels)
current_final_bboxes += [temp_bboxes]
self.recent_bboxes[-1] = current_final_bboxes
self.processed_image = np.copy(img)
for i in range(nrounds):
heat = np.zeros_like(img[:,:,0]).astype(np.float)
# We only count heat map for same scale of bboxes.
weight = 0
for bboxes_in_one_frame in self.recent_bboxes:
weight += float(1/len(self.recent_bboxes))
heat = add_heat(heat,bboxes_in_one_frame[i], weight)
# Make sure first several frames will not have too big treshold
heat_current = apply_threshold(heat, max(len(self.recent_bboxes)/2, 2))
heatmap_current = np.clip(heat_current, 0, 255)
labels = label(heatmap_current)
self.processed_image, _ = draw_labeled_bboxes(self.processed_image, labels)
Run for video¶
In [11]:
from moviepy.editor import VideoFileClip
cars_detector = CarsDetector()
def process_video_image(img):
cars_detector.detect_car(img)
return cars_detector.processed_image
# The following
#video_input = './project_video.mp4'
#video_output = './test_clip2.mp4'
#clip1 = VideoFileClip(video_input)
#video_clip = clip1.subclip(25, 30)
#video_clip.write_videofile(video_output, audio=False)
# Generate test images for tuning use.
#for i in range(20):
# filename = './video_frame_' + str(i) + '.jpg'
# time = 45 + float(i/10)
# clip1.save_frame(filename, t=time)
In [12]:
cars_detector = CarsDetector()
for i in range(20):
filename = './video_frame_' + str(i) + '.jpg'
img = cv2.imread(filename)
cars_detector.detect_car(img)
f, ax = plt.subplots(1, 1, figsize=(25,25))
ax.imshow(cv2.cvtColor(cars_detector.processed_image, cv2.COLOR_BGR2RGB))
plt.subplots_adjust(wspace=0, hspace=0)
